Projects and Content


Tutorial: Computer Vision at Scale with Dask and PyTorch

I use the Stanford Dogs dataset to demonstrate accelerating an image classification problem with GPU Clusters. If you have been thinking about GPUs but don’t know where to start, or what they might be good for, I recommend this as a place to start!

Keywords: machine learning, python, deep learning




Article: Make Your Data Move: Using Schedulers with Data Storage to Generate Business Value

In concert with my presentation at ODSC Europe 2020 I wrote a blog post to discuss why and how you might use scheduled jobs to make your data infrastructure work better for modeling and machine learning.

Keywords: job scheduling, airflow, data warehousing




Model Behavior Video

I appeared in a video series for Uptake about how modeling industrial failures work - this was really fun, and I hope people will give it a look! I talk about a specific failure of diesel locomotive hardware, and how I solved it while I was working at Uptake.




Radlibs! in R or in Python

I wrote a silly R package called radlibs that allows you to make your own madlibs. Then I wrote a version in Python. Then I added them to CRAN and pypi. Data science doesn’t always have to be serious. Use install.packages("radlibs") or pip install radlibs to get these packages. Issues and feedback welcome!

Keywords: python, r, packages




Evaluation of R Forwards Package Workshop

I recently co-taught a daylong course for a group of 30 women/gender nonbinary students about how to write R packages- we had a really good time! I analyzed our pre- and post- surveys in a notebook, to check how effective the day was for students.

Keywords: r, data visualization




Kiva Loan Data Analysis

  • GIS analysis of loans by country with attention to economic conditions in countries
  • Drilldown on some of the thematic areas of the loans
  • Data munging of the regional data provided
  • Exploring modeling potential- if repayment time can be predicted, or anomaly detection if labeling outcomes is not possible

Keywords: r, data visualization




Fun with Real Estate Data

This project is a kaggle kernel, in which I walked the reader through the process of cleaning and modeling the data from a real estate prices dataset, using linear modeling, random forests, and gradient boosting (xgboost). My most popular kernel to date! This one also produced respectable competition results, and was chosen for special recognition by the Kaggle admins. (I won a mug!)

Update: Read the interview I did regarding this project (and the other fabulous winners)! http://blog.kaggle.com/2017/03/29/predicting-house-prices-playground-competition-winning-kernels

Keywords: machine learning, data cleaning